CCminer(SP-MOD) Modded GPU kernels.

[sp_]

The neoscrypt hashrate can easily be verified. Why don`t you build it yourself with cuda 7.5.
A 750ti will do 60-70 khash with your code.
And 90-100khash with my code.

[scryptr]

If you want to test the latest @git the cryptomining blog has just compiled a new version. release 77+

http://cryptomining-blog.com/6341-updated-windows-binary-of-the-ccminer-1-5-77-git-fork-by-sp-for-maxwell/

DJM34's Neoscrypt kernal is hurting when compiled with cuda 7.5.

I managed to increase it 50% faster on the 750ti with some tuning.  (from 60KHASH to 90KHASH). (release 74 cuda 6.5 is doing 180khash)

After changing the compiler from 6.5 to 7.5, all the kernals has to be retuned/rewritten. This takes time, but is Necessary.

HYBRID COMPILED RELEASE--

Pallas described a method to compile CCminer with both CUDA 7.5 and 6.5.  The idea was that the Neoscrypt could be compiled with CUDA 6.5, and all other algos with CUDA 7.5.  The CUDA 7.5 compilation would be first, and then a re-compile with CUDA 6.5 would compile only the Neoscrypt algo.  The hybrid would work until Neoscrypt code is re-tuned.

For a brief moment, I thought that DJM34 might contibute with some Nescrypt updates.  I'd contribute to both you and DJM34, I don't know why the whole thing became "Wreck The Halls" with Yuletide folly.  

I'm pretty poor on my coding skills.  But, I think most miners would chip in for even a simple hybrid.  If DJM34 wants some well-deserved contributions, maybe he could add some new Neoscrypt code.


--scryptr

EDIT:  The hybrid release would allow the use of auto-switching software with only one mining executable.      --scryptr

[djm34]

If you want to test the latest @git the cryptomining blog has just compiled a new version. release 77+

http://cryptomining-blog.com/6341-updated-windows-binary-of-the-ccminer-1-5-77-git-fork-by-sp-for-maxwell/

DJM34's Neoscrypt kernal is hurting when compiled with cuda 7.5.

I managed to increase it 50% faster on the 750ti with some tuning.  (from 60KHASH to 90KHASH). (release 74 cuda 6.5 is doing 180khash)

After changing the compiler from 6.5 to 7.5, all the kernals has to be retuned/rewritten. This takes time, but is Necessary.

HYBRID COMPILED RELEASE--

Pallas described a method to compile CCminer with both CUDA 7.5 and 6.5.  The idea was that the Neoscrypt could be compiled with CUDA 6.5, and all other algos with CUDA 7.5.  The CUDA 7.5 compilation would be first, and then a re-compile with CUDA 6.5 would compile only the Neoscrypt algo.  The hybrid would work until Neoscrypt code is re-tuned.

For a brief moment, I thought that DJM34 might contibute with some Nescrypt updates.  I'd contribute to both you and DJM34, I don't know why the whole thing became "Wreck The Halls" with Yuletide folly. 
--scryptr

lolwut ? missing the reference 

I said I might, but I don't know if I will have time... will keep you inform

[scryptr]

If you want to test the latest @git the cryptomining blog has just compiled a new version. release 77+

http://cryptomining-blog.com/6341-updated-windows-binary-of-the-ccminer-1-5-77-git-fork-by-sp-for-maxwell/

DJM34's Neoscrypt kernal is hurting when compiled with cuda 7.5.

I managed to increase it 50% faster on the 750ti with some tuning.  (from 60KHASH to 90KHASH). (release 74 cuda 6.5 is doing 180khash)

After changing the compiler from 6.5 to 7.5, all the kernals has to be retuned/rewritten. This takes time, but is Necessary.

HYBRID COMPILED RELEASE--

Pallas described a method to compile CCminer with both CUDA 7.5 and 6.5.  The idea was that the Neoscrypt could be compiled with CUDA 6.5, and all other algos with CUDA 7.5.  The CUDA 7.5 compilation would be first, and then a re-compile with CUDA 6.5 would compile only the Neoscrypt algo.  The hybrid would work until Neoscrypt code is re-tuned.

For a brief moment, I thought that DJM34 might contibute with some Nescrypt updates.  I'd contribute to both you and DJM34, I don't know why the whole thing became "Wreck The Halls" with Yuletide folly.  
--scryptr

lolwut ? missing the reference  

I said I might, but I don't know if I will have time... will keep you inform

DECK THE HALLS WITH YULETIDE FOLLY--

Falalalala, lalalala!  English Christmas carol, widely sung in the UK, USA, Canada, etc.  It refers to Christmas decorations about the many households (Not bloody red noses!   ).   --scryptr

EDIT:  I think the original line was "Deck the halls with boughs of holly!"  --scryptr

[djm34]

If you want to test the latest @git the cryptomining blog has just compiled a new version. release 77+

http://cryptomining-blog.com/6341-updated-windows-binary-of-the-ccminer-1-5-77-git-fork-by-sp-for-maxwell/

DJM34's Neoscrypt kernal is hurting when compiled with cuda 7.5.

I managed to increase it 50% faster on the 750ti with some tuning.  (from 60KHASH to 90KHASH). (release 74 cuda 6.5 is doing 180khash)

After changing the compiler from 6.5 to 7.5, all the kernals has to be retuned/rewritten. This takes time, but is Necessary.

HYBRID COMPILED RELEASE--

Pallas described a method to compile CCminer with both CUDA 7.5 and 6.5.  The idea was that the Neoscrypt could be compiled with CUDA 6.5, and all other algos with CUDA 7.5.  The CUDA 7.5 compilation would be first, and then a re-compile with CUDA 6.5 would compile only the Neoscrypt algo.  The hybrid would work until Neoscrypt code is re-tuned.

For a brief moment, I thought that DJM34 might contibute with some Nescrypt updates.  I'd contribute to both you and DJM34, I don't know why the whole thing became "Wreck The Halls" with Yuletide folly. 
--scryptr

lolwut ? missing the reference 

I said I might, but I don't know if I will have time... will keep you inform

DECK THE HALLS WITH YULETIDE FOLLY--

Falalalala, lalalala!  English Christmas carol, widely sung in the UK, USA, Canada, etc.  It refers to Christmas decorations about the many households (Not bloody red noses!  ).   --scryptr

EDIT:  I think the original line was "Deck the halls with boughs of holly!"  --scryptr

found it 

[lmaonade80]

So... Must I install CUDA 6.5 for faster NeoScrypt mining? And if so, would the other algorithms suffer as a result?

[djm34]

So... Must I install CUDA 6.5 for faster NeoScrypt mining? And if so, would the other algorithms suffer as a result?

Install Cuda 6.5 you must, and "normal" neoscrypt speed, you'll get

Master Yoda

[joblo]

So... Must I install CUDA 6.5 for faster NeoScrypt mining? And if so, would the other algorithms suffer as a result?

You must compile ccminer with cuda 6.5...

Other algos will be affected but in unpredictable ways.

[scryptr]

WINDOWS RELEASE VERSIONS--

If you use release dot 74, as compiled by sp_ , you will get optimal Neoscrypt rates.  Release dot 74 was the last release compiled by sp_ that used CUDA 6.5.

If you use release dot 77+, it is compiled with CUDA 7.5, and Neoscrypt rates are lower.  Other algos have been improved.

In Linux, it is possible to compile a hybrid version if you use both CUDA 6.5 and 7.5, as described by Pallas a few pages back in the thread.  At least, he describes the hybrid method.  No programmer has released a hybrid pre-compiled version.

--scryptr

[joblo]

WINDOWS RELEASE VERSIONS--

If you use release dot 74, as compiled by sp_ , you will get optimal Neoscrypt rates.  Release dot 74 was the last release compiled by sp_ that used CUDA 6.5.

If you use release dot 77+, it is compiled with CUDA 7.5, and Neoscrypt rates are lower.  Other algos have been improved.

In Linux, it is possible to compile a hybrid version if you use both CUDA 6.5 and 7.5, as described by Pallas a few pages back in the thread.  At least, he describes the hybrid method.  No programmer has released a hybrid pre-compiled version.

--scryptr

And just for completeness, if you have SM3.5 cards (ie 780ti) the old neoscrypt kernel in ccminer 1.5.58 is 15% faster.
All this makes for a messy recipe but that's what you get with optimizing. One size does not fit all.

[lmaonade80]

WINDOWS RELEASE VERSIONS--

If you use release dot 74, as compiled by sp_ , you will get optimal Neoscrypt rates.  Release dot 74 was the last release compiled by sp_ that used CUDA 6.5.

If you use release dot 77+, it is compiled with CUDA 7.5, and Neoscrypt rates are lower.  Other algos have been improved.

In Linux, it is possible to compile a hybrid version if you use both CUDA 6.5 and 7.5, as described by Pallas a few pages back in the thread.  At least, he describes the hybrid method.  No programmer has released a hybrid pre-compiled version.

--scryptr

Awesome, thanks. I was able to bump up my cards. Much appreciated.

[djm34]

WINDOWS RELEASE VERSIONS--

If you use release dot 74, as compiled by sp_ , you will get optimal Neoscrypt rates.  Release dot 74 was the last release compiled by sp_ that used CUDA 6.5.

If you use release dot 77+, it is compiled with CUDA 7.5, and Neoscrypt rates are lower.  Other algos have been improved.

In Linux, it is possible to compile a hybrid version if you use both CUDA 6.5 and 7.5, as described by Pallas a few pages back in the thread.  At least, he describes the hybrid method.  No programmer has released a hybrid pre-compiled version.

--scryptr

is it so difficult for you people to have several versions on your computers ? (while Christian Büchner was still working on cudaminer and ccminer I was having something like 4 to 5 concurrent version of it...)  

[antantti]

is it so difficult for you people to have several versions on your computers ? (while Christian Büchner was still working on cudaminer and ccminer I was having something like 4 to 5 concurrent version of it...)  

And I was about to donate.

[Eliovp]

is it so difficult for you people to have several versions on your computers ? (while Christian Büchner was still working on cudaminer and ccminer I was having something like 4 to 5 concurrent version of it...)  

And I was about to donate.

I lol'ed.. Christmas Spirit yet again :p

[scryptr]

WINDOWS RELEASE VERSIONS--

If you use release dot 74, as compiled by sp_ , you will get optimal Neoscrypt rates.  Release dot 74 was the last release compiled by sp_ that used CUDA 6.5.

If you use release dot 77+, it is compiled with CUDA 7.5, and Neoscrypt rates are lower.  Other algos have been improved.

In Linux, it is possible to compile a hybrid version if you use both CUDA 6.5 and 7.5, as described by Pallas a few pages back in the thread.  At least, he describes the hybrid method.  No programmer has released a hybrid pre-compiled version.

--scryptr

is it so difficult for you people to have several versions on your computers ? (while Christian Büchner was still working on cudaminer and ccminer I was having something like 4 to 5 concurrent version of it...)  

ALGO-SWITCHING SOFTWARE--

It is convenient to have a single version for algo-switching software.  Not absolutely required, it just is easier for the miner.  I haven't been able to work the bug's out of Skunk's outdated BASH script for algo-switching, he did use different miniing software for different algos.  It would be nice to have a compiled C-language Linux executable that works like NiceHash Miner.  NiceHash miner uses dotNet code, and that is Windows stuff.  I tried to implement dotNet emulation in Linux, it didn't go well. 

--scryptr

[chrysophylax]

WINDOWS RELEASE VERSIONS--

If you use release dot 74, as compiled by sp_ , you will get optimal Neoscrypt rates.  Release dot 74 was the last release compiled by sp_ that used CUDA 6.5.

If you use release dot 77+, it is compiled with CUDA 7.5, and Neoscrypt rates are lower.  Other algos have been improved.

In Linux, it is possible to compile a hybrid version if you use both CUDA 6.5 and 7.5, as described by Pallas a few pages back in the thread.  At least, he describes the hybrid method.  No programmer has released a hybrid pre-compiled version.

--scryptr

is it so difficult for you people to have several versions on your computers ? (while Christian Büchner was still working on cudaminer and ccminer I was having something like 4 to 5 concurrent version of it...)  

ALGO-SWITCHING SOFTWARE--

It is convenient to have a single version for algo-switching software.  Not absolutely required, it just is easier for the miner.  I haven't been able to work the bug's out of Skunk's outdated BASH script for algo-switching, he did use different miniing software for different algos.  It would be nice to have a compiled C-language Linux executable that works like NiceHash Miner.  NiceHash miner uses dotNet code, and that is Windows stuff.  I tried to implement dotNet emulation in Linux, it didn't go well. 

--scryptr

do you have a git - or copy - of the script? ... i have a copy of your older one - and needed some work on it - but just havent had time ...

and ill be going away for a christmas break for a few days - so i wont be back till new years ...

i will pop in and out occasionally to check on the threads - but wont be working much on granite - or thefarm till i get back ... a few plans already started and a few more in the making ... 2016 will be a nice prosperous year ...

happy christmas guys ...

#crysx

[grendel25]

... for nvidia there is someone which doesn't do any work and ask people to send him the donation instead to the real developer...

I just gave you 5% more quark on the gtx 970 and 5% more lyra2v2 on the gtx 950 and gtx 960.

The 5% lyra2v2 increase have given me 0 BTC in donations and I've spent many hours.

The lyra2v2 optimalizations doesn't work on the 750ti and the gtx 980/980ti cards.


Here are my last commits:

https://github.com/sp-hash/ccminer/commits/windows

The latest changeset are mostly compiler adjustments/tuning to get the cuda 7.5 compiler to perform as fast as cuda 6.5.

Hang in there sp_  http://www.reuters.com/article/us-global-markets-bitcoin-analysis-idUSKBN0U60GM20151223?feedType=RSS&feedName=businessNews

[grendel25]

Is the 750Ti still the best bang for the buck?

[scryptr]

Is the 750Ti still the best bang for the buck?

YES--

But there is an argument for the newer GTX 9XX cards because the cost for the rig(s) is less.  Fewer motherboards and power supplies are required for the same total hash level.  A 750ti by itself is the best bang for the buck on card price and power cost.

--scryptr

[grendel25]

Is the 750Ti still the best bang for the buck?

YES--

But there is an argument for the newer GTX 9XX cards because the cost for the rig(s) is less.  Fewer motherboards and power supplies are required for the same total hash level.  A 750ti by itself is the best bang for the buck on card price and power cost.

--scryptr

Copy, thank you.